June 22, 2018
Atlas Co-Founder and Partner, Jason Shepherd was recently quoted in Real Estate Business Institute’s online magazine discussing listing prices. Addressing the frequent valuation discrepancies of larger websites like Zillow, Shepherd says:
“To combat the pricing metrics of websites like Zillow or Realtor.com, it’s important to describe the shortfalls of algorithm-based valuation in real estate. The accuracy of housing value websites depends on the dataset, algorithms, and machine learning.
Artificial intelligence is only as good as the data it’s given, and with real estate, there’s not enough aggregated data for the algorithms to produce an output with a good confidence interval. It’s still very early in its data lifecycle, so it isn’t close to perfect, but it’s getting better.
The current issue with AI in real estate is the infrequency of data points. Real estate doesn’t transact as often as other forms of data-driven businesses. For example, stocks trade value every fraction of a second, and online poker bots can see hundreds of hands at a time. In these examples, the robust datasets allow a machine to learn and then determine valuation. The average homeowner buying or selling their primary residence once every seven years doesn’t allow for the most accurate valuation logic.
The even larger issue for accurate data resides in such housing variables as the interior condition, basement finish, views, whether there’s a liquor store next door, surrounding traffic, five pitbulls in the neighbor’s backyard, excessive HOA fees, and the list goes on. These variables can be so specific to a particular home that even the house next door or the house across the street might not be susceptible to these types of valuation adjustments.
These predictive valuation websites will continue to get better as a natural product of more data and information. At this time, they should be used as a starting point but never to determine list price.”